Research On Feature Gene Extraction Method Based On The Dynamical Neighborhood

With the development of DNA microarray technology, it is so easy toobtain thousands of the biomedical data in form of gene expressionprofiles quickly. For this reason, the data have many characteristics suchas small sample size, high dimension, high noise, multiple correlation,nonlinear and uneven distribution and so on. The characteristics makedifficult to understand and analyze the biomedical data. Fortunately, thisproblem can be solved by selecting the relevant features or extracting theessential features from the original data, where the former methodology isbelonging to feature selection and the latter is just feature extraction. Butneither of them is comprehensively consider the characteristics of the genedata. So the research direction of the article is to propose a dimensionreduction method which is suitable for the gene data.The theme of the paper is full considering the nonlinear, multiplecorrelation, and uneven distribution characteristics of gene data based ondynamical neighborhood. Two feature gene extraction methods have been proposed. The main research work is as follows:Firstly, a feature gene extraction method based on dynamicalneighborhood and locally linear discriminant embedding (LLDE) isproposed in this paper, named as dynamical locally linear discriminantEmbedding (DLLDE). The method fully considers the nonlinear anduneven distribution characteristics of the gene data. The proposed methodeffectively combines LLDE and dynamical neighborhood, and selects theKNN as the prediction classification. We apply it to the three publicdatasets. The results of our experiments show that our method is moreaccurate and stable than LLDE.Secondly, dimensionality reduction is more important than theclassification in the analysis of gene microarray data. On account of thecharacteristics of the gene data, dimensionality reduction is very difficultfor the gene data. In order to solve the problems, a new dimensionalityreduction algorithm is proposed for feature gene extraction, whichcontains the random forest (RF), dynamical neighborhood and LLDE.Different from LLDE, the proposed method not only keeps the advantages of LLDE, but also takes into account the multiple correlations among thegenes and the combined effects of all the genes and uneven distribution.The results of our experiments show that our method maintains highaccuracy and is more stable than other methods.